salad package¶

Test salad is a toolbox for semi-supervised adaptive learning across domains. It is mainly meant for domain adaptation, semi-supervised learning and transfer learning in general, using PyTorch as a backend.

$$ mathbb{E}_{p_t} [ mathcal{L}(x_t, y_t) ] $$

Subpackages¶

Submodules¶

salad.optim module¶

class salad.optim.JointOptimizer(*optims)¶

Bases: object

Concat multiple optimizers

Parameters:	*optims (list of `torch.optim.Optimizer`) – Optimizers. The `step` and `zero_grad` functions will be executed in the same order.

step()¶

zero_grad()¶

class salad.optim.WeightEMA(params, src_params, alpha=0.999)¶

Bases: object

Exponential moving average weight optimizer for mean teacher model

Used for Self-Ensembling, code adapted from [1].

salad.structural module¶

Helper functions for structural learning

class salad.structural.CompressedResnet(backbone)¶

Bases: torch.nn.modules.module.Module

ResNet Variant where the batch norm statistics are merged into the transformation matrices

forward(x)¶

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class salad.structural.FixedBottleneck(conv, downsample)¶